Evaluating Forest Disturbance Detection Methods based on Satellite Image Time Series for Amazon Deforestation Alerts

Abstract. This study explores automated detection methods of forest disturbances using satellite image time series for Amazon deforestation alerts. The research focuses on two municipalities in southern Amazonas, Brazil, known for high numbers of deforestation alerts. Five methods—BFAST Monitor, CCDC, COLD, SCCD, and LSTM—were applied to Landsat image time series from 2017 to 2020 to identify forest disturbances and their effectiveness were evaluated, by comparing their results with alerts from the Brazilian Real-time Deforestation Detection System (DETER). The results demonstrate that the COLD and SCCD methods achieved the highest concordance rates with DETER alerts, at 82% and 85%, respectively, indicating their superior performance in disturbance detection. The LSTM method also performed well, with an 83% concordance rate, showcasing the potential of deep learning techniques in satellite image time series. The CCDC method followed with a 75% concordance rate, and the BFAST method had a concordance rate of 72%. This study highlights the importance of utilizing advanced modeling techniques and multi-spectral analysis for effective forest disturbance detection. The results underscore the need for continued refinement and calibration of these methods to enhance their precision and reliability.